Skip to contents

multixoR 0.2.1

  • Clarify the default MCTS score, half-credit draw convention, shipped heuristic calibrator and limits of its decisive-outcome training fit. Calibration diagnostics describe fitting data, not held-out performance.

  • Reject missing, fractional, nonnumeric and out-of-range player values in mxo_evaluate() and mxo_win_prob() before coercion or terminal-state returns. Valid player perspectives and numerical results are unchanged.

multixoR 0.2.0

CRAN preparation

  • Maintainer address changed to .
  • .Rbuildignore covers the development folders, so the source package no longer contains an empty tools/ directory; README links to the rules specification on GitHub.
  • Added spelling WORDLIST entries and updated cran-comments.md.
  • The three slowest tests (opening tables and an isotonic calibration fit, about two minutes together) are skipped on CRAN.

Bug fixes and improvements

  • Make human play functional in the Shiny controls, honor initial difficulty, add AI replies and explicit AI stepping, and check the required plotly dependency.

  • Forward named AI/rating engine arguments, including explicit NULL budgets.

  • Bound hard AI with cooperative deadline checks through tree traversal and rollouts; use random rollouts for MCTS move ratings by default.

  • Rank search ratings by search value and leave unsupported probability labels missing. Reject invalid or fractional search depths before conversion.

  • Store distinct isotonic knots and handle constant-score calibration data.

multixoR 0.1.0

Initial release.

Engine

AI and evaluation

  • mxo_evaluate() — heuristic weighted m-line score with a cross-timeline multiplier (w_timeline).
  • mxo_search() — negamax + α-β with branch_policy (all / promising / none).
  • mxo_mcts() — light UCT with random / heuristic rollouts, time_budget, seedable.
  • mxo_win_prob() — heuristic logistic via a fitted internal calibrator (see Calibration below) or MCTS visit aggregation.
  • mxo_rate_moves() — type-stable per-move tibble with rank + blunder/strong/best label.
  • mxo_ai_move() — easy / medium / hard difficulty knob.

Simulation and strategy

Visualisation

App and docs

  • mxo_run_app() launches the bundled Shiny app (shiny, bslib, DT in Suggests) with the multiverse / board / threats / tree / analysis tabs.
  • mxo_example_game() deterministic short game with a branch, used by docs / vignettes / the app default state.
  • Four vignettes: Getting started, Rules and geometry, Analysis, Self-play and simulation.
  • pkgdown site at https://r-heller.github.io/multixoR/ themed via the CTTIR suite’s themakR template.
  • GitHub Actions: R-CMD-check matrix (macOS / Windows / Ubuntu × release + devel + oldrel-1), pkgdown deploy, codecov.

Performance

  • Pure-R engine for v1.0. The hot loops in .mxo_line_features() and .mxo_check_win_at() are deliberately Rcpp-ready so a future compiled backend is a drop-in.